CN109215034A - A kind of Weakly supervised image, semantic dividing method for covering pond based on spatial pyramid - Google Patents

A kind of Weakly supervised image, semantic dividing method for covering pond based on spatial pyramid Download PDF

Info

Publication number
CN109215034A
CN109215034A CN201810739297.1A CN201810739297A CN109215034A CN 109215034 A CN109215034 A CN 109215034A CN 201810739297 A CN201810739297 A CN 201810739297A CN 109215034 A CN109215034 A CN 109215034A
Authority
CN
China
Prior art keywords
pond
pyramid
semantic
spatial pyramid
characteristic pattern
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201810739297.1A
Other languages
Chinese (zh)
Other versions
CN109215034B (en
Inventor
朱策
段昶
文宏雕
徐榕健
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chengdu Map Technology Co Ltd
Original Assignee
Chengdu Map Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Chengdu Map Technology Co Ltd filed Critical Chengdu Map Technology Co Ltd
Priority to CN201810739297.1A priority Critical patent/CN109215034B/en
Publication of CN109215034A publication Critical patent/CN109215034A/en
Application granted granted Critical
Publication of CN109215034B publication Critical patent/CN109215034B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a kind of Weakly supervised image, semantic dividing methods that pond is covered based on spatial pyramid, comprising the following steps: selectes a convolutional neural networks H, handles input picture X by convolutional neural networks H, obtain characteristic of division figure；Spatial pyramid pond module is established according to characteristic of division figure, then carries out spatial pyramid cover, obtains output characteristic pattern；Classification is calculated according to output characteristic pattern and activates vector sum class probability vector, is then established competitive spatial pyramid and is covered pond loss function；Pond loss function training convolutional neural networks H is covered according to competitive spatial pyramid and extracts segmentation characteristic pattern.The present invention realizes richer local feature, the provincial characteristics more perfect and target sizes of excavation and the more robust Weakly supervised image, semantic parted pattern of posture, the extractability for improving local semantic information strengthens localized target or the recognition capability at position in Weakly supervised semantic segmentation.

Description

A kind of Weakly supervised image, semantic dividing method for covering pond based on spatial pyramid
Technical field
The invention belongs to technical field of computer vision, and in particular to a kind of weak prison that pond is covered based on spatial pyramid Superintend and direct image, semantic dividing method.
Background technique
Image, semantic is divided as basic Computer Vision Task, and target is to be divided all pixels in image Class.Since it can be fully understood by image in Pixel-level, so all for other visual tasks such as image classification and target identification There is booster action.But because the production of Pixel-level label data needs to expend great effort, the image, semantic of full monitor mode Segmentation is difficult to fast implement extensive extension, and the Weakly supervised image, semantic dividing method for then relying on image class label is extensive Research.
In past computer vision research, there are many pyramid model application.Famous SIFT algorithm passes through La Pula This pyramid extracts key point Expressive Features, becomes a kind of detective operators with Scale invariant property.Later, pyramid model It is successfully applied to convolutional neural networks.(spatial pyramid pooling, SPP) be used to roll up in spatial pyramid pond Local features are extracted in product neural network tail portion, and SPP obtains preferably effect in image classification and target identification.Cavity Spatial pyramid (atrous spatial pyramid pooling, ASPP) module is based on multiple dispersion ratio (dilated Rate empty convolution (dilated convolution)) obtains pyramid feature, and is applied to image, semantic and divides.Golden word Characteristic pattern is sliced into different spaces and explores part and whole semantic relation by tower scene cut model, by entirety and part The semantic information in region compares and combines to obtain more robust segmentation result.
There are no relatively successfully applications in Weakly supervised semantic segmentation for pyramid model.On the one hand, with global pool knot The sorter network of tail is only capable of obtaining target-recognition region due to learning in a manner of Weakly supervised, lacks mentioning for local region information Ability is taken, the shortcomings that pyramid model obviously can make up global pool due to the characteristic of its own.On the other hand, pyramid mould Although type has the characteristics of multi-scale information and local message merge, but how preferably on the basis of grasping global information The problem that time important semantic information is excavated in subregion is fully solved not yet.
Summary of the invention
It is an object of the invention to: above-mentioned deficiency in the prior art is solved, is provided a kind of based on spatial pyramid cover The Weakly supervised image, semantic dividing method in pond, improves the extractability of local semantic information, strengthens Weakly supervised semantic point Cut the recognition capability at middle localized target or position.
To achieve the goals above, the technical solution adopted by the present invention are as follows:
A kind of Weakly supervised image, semantic dividing method for covering pond based on spatial pyramid, comprising the following steps:
Step 1: selecting a convolutional neural networks H, handles input picture X by convolutional neural networks H, is classified Characteristic pattern；
Step 2: establishing spatial pyramid pond module according to characteristic of division figure, then carries out spatial pyramid cover, obtains To output characteristic pattern；
Step 3: calculating classification according to output characteristic pattern and activate vector sum class probability vector, then establishes competitive empty Between pyramid cover pond loss function；
Step 4: it covers pond loss function training convolutional neural networks H according to competitive spatial pyramid and extracts and divide Cut characteristic pattern.
Further, the step of input picture X being handled by convolutional neural networks H in above-mentioned steps one specifically:
Step 101: selecting a convolutional neural networks H, map to obtain segmentation characteristic pattern by the convolutional neural networks HWherein, C is target category quantity, and W is weight parameter, For real number field；
Step 102: characteristic pattern dimensionality reduction will be divided by 1 × 1 convolution, obtain characteristic of division figureIt calculates public Formula are as follows:
Fcls=Fseg×W+b
Wherein, b is biasing,
Further, above-mentioned step two specifically:
Step 201: average pond being carried out to all characteristic of division figures, while specifying total pyramid number and Chi Huahe big It is small, obtain corresponding pyramid pond characteristic pattern Pi, wherein i is pyramid serial number, and value range arrives total pyramid number between 1 Between mesh, i-th of pyramid pond characteristic pattern is calculatedCalculation formula are as follows:
Wherein, kiFor i-th of pyramidal pond core size, x is characterized figure abscissa, and y is characterized figure ordinate, △ kx For pond core abscissa, △ kyFor pond core ordinate, c is semantic classes quantity；
Step 202: each pyramid pond characteristic pattern being generated and covers tensorRegion is concurrently set to cover Lid ratio and cover quantization parameter；
Step 203: according to the output characteristic pattern for covering tensor computation spatial pyramid cover pondMeter Calculate formula are as follows:
Oi=Pi⊙Mi
Wherein, ⊙ indicates hadamard product.
Further, in above-mentioned steps 202, remember that i-th of pyramid covers tensor and beRemember jth eka-gold Word tower covers tensor, the region that the different classes of pyramid covers tensor covers position and covers quantization Coefficient is consistent:
The ratio that the cover region generated at random accounts for the pyramid pond number of regions is τ,
Wherein, the value range of τ is [0,1], and the value of τ is bigger, and the region for indicating to cover is more,It indicates (1-τ)HiThe value that W is obtained is rounded downwards, and I () indicates that indicator function, I () take 1 when input condition meets, other situations Under take 0, i.e.,The position of satisfaction is to be retained region, and other situations are the region covered；
The cover quantization parameter ρ is for inhibiting signal；
Wherein, the value range of ρ is [0,1].
Further, classification is calculated in above-mentioned steps three activate vector sum class probability vector specifically:
Step 301: calculating classification and activate vector oi, i.e., the output characteristic pattern that spatial pyramid covers pond is classified as one Value, calculation formula are as follows:
Step 302: class probability vector being calculated by Softmax function, obtains the probability of each semantic classes to be determined Value, calculation formula are as follows:
Further, in above-mentioned steps three, establish competitive spatial pyramid cover pond loss function specifically include with Lower step:
Step 303: calculating the Classification Loss l of each pyramid gradei, calculation formula are as follows:
Wherein,1 other places are taken to take 0, T for image, semantic class label vector and at the corresponding semantic classes target of appearance For transposition symbol；
Step 304: the 0th grade of global pool information being fully retained, other pyramids for having carried out information cover etc. are grading Mobile state competition, calculation formula are as follows:
Wherein, lclsFor loss function of always classifying, N is pyramid number of levels, l0For the damage of global pool pyramid grade It loses；
Further, above-mentioned step four specifically: error is calculated with loss function of always classifying and is calculated by backpropagation Method adjusts network parameter after the opposite variation of the loss function is less than 5%, directly by FsegIt takes out and is taken in classification dimension Prediction segmentation figure can be obtained in argmax
Y=argmax (Fseg)
Wherein, argmax FsegThe 3rd dimension.
By adopting the above-described technical solution, the beneficial effects of the present invention are:
It is more robust that the present invention realizes richer local feature, the provincial characteristics more perfect and target sizes of excavation and posture Weakly supervised image, semantic parted pattern.
The present invention obtains finer segmentation result to preferably excavate local message, to spatial pyramid pond module Increase cover mechanism, in addition to the diversity of extension feature figure, identification region can also be inhibited and then encourage identification more times The study in region；
The case where misplacing just present invention is alternatively directed to different pyramidal cover regions, proposes that auxiliary space pyramid is covered The competitive spatial pyramid of pondization training covers pond loss function, reduces the risk of network training failure；
Global pool information of the invention make network to the scale of target will not tetchiness, and spatial pyramid is covered Gai Chihua more has versatility in other visual tasks.
Detailed description of the invention
Fig. 1 is the Weakly supervised image, semantic parted pattern schematic diagram of the invention that pond is covered based on spatial pyramid.
Fig. 2 is that spatial pyramid of the invention covers pond structural schematic diagram.
Fig. 3 is that the Weakly supervised image, semantic of the invention for covering pond based on spatial pyramid divides flow diagram.
Fig. 4 is segmentation result contrast schematic diagram of the invention.
Specific embodiment
Referring to attached drawing 1-4, specific description is done to embodiments of the present invention.
A kind of Weakly supervised image, semantic dividing method for covering pond based on spatial pyramid, comprising the following steps:
Step 1: selecting a convolutional neural networks H, handles input picture X by convolutional neural networks H, is classified Characteristic pattern；
Step 2: establishing spatial pyramid pond module according to characteristic of division figure, then carries out spatial pyramid cover, obtains To output characteristic pattern；
Step 3: calculating classification according to output characteristic pattern and activate vector sum class probability vector, then establishes competitive empty Between pyramid cover pond loss function；
Step 4: it covers pond loss function training convolutional neural networks H according to competitive spatial pyramid and extracts and divide Cut characteristic pattern.
Further, the step of input picture X being handled by convolutional neural networks H in above-mentioned steps one specifically:
Step 101: selecting a convolutional neural networks H, map to obtain segmentation characteristic pattern by the convolutional neural networks HWherein, C is target category quantity, and W is weight parameter, For real number field；
Step 102: characteristic pattern dimensionality reduction will be divided by 1 × 1 convolution, obtain characteristic of division figureIt calculates public Formula are as follows:
Fcls=Fseg×W+b
Wherein, b is biasing,
Further, above-mentioned step two specifically:
Step 201: average pond being carried out to all characteristic of division figures, while specifying total pyramid number and Chi Huahe big It is small, obtain corresponding pyramid pond characteristic pattern Pi, wherein i is pyramid serial number, and value range arrives total pyramid number between 1 Between mesh, i-th of pyramid pond characteristic pattern is calculatedCalculation formula are as follows:
Wherein, kiFor i-th of pyramidal pond core size, x is characterized figure abscissa, and y is characterized figure ordinate, △ kx For pond core abscissa, △ kyFor pond core ordinate, c is semantic classes quantity；
Step 202: each pyramid pond characteristic pattern being generated and covers tensorRegion is concurrently set to cover Lid ratio and cover quantization parameter；
Step 203: according to the output characteristic pattern for covering tensor computation spatial pyramid cover pondMeter Calculate formula are as follows:
Oi=Pi⊙Mi
Wherein, ⊙ indicates hadamard product.
Further, in above-mentioned steps 202, remember that i-th of pyramid covers tensor and beRemember jth eka-gold Word tower covers tensor, the region that the different classes of pyramid covers tensor covers position and covers quantization Coefficient is consistent:
The ratio that the cover region generated at random accounts for the pyramid pond number of regions is τ,
Wherein, the value range of τ is [0,1], and the value of τ is bigger, and the region for indicating to cover is more,It indicates (1-τ)HiThe value that W is obtained is rounded downwards, and I () indicates that indicator function, I () take 1 when input condition meets, other situations Under take 0, i.e.,The position of satisfaction is to be retained region, and other situations are the region covered；
The cover quantization parameter ρ is for inhibiting signal；
Wherein, the value range of ρ is [0,1].
Further, classification is calculated in above-mentioned steps three activate vector sum class probability vector specifically:
Step 301: calculating classification and activate vector oi, i.e., the output characteristic pattern that spatial pyramid covers pond is classified as one Value, calculation formula are as follows:
Step 302: class probability vector being calculated by Softmax function, obtains the probability of each semantic classes to be determined Value, calculation formula are as follows:
Further, in above-mentioned steps three, establish competitive spatial pyramid cover pond loss function specifically include with Lower step:
Step 303: calculating the Classification Loss l of each pyramid gradei, calculation formula are as follows:
Wherein,1 other places are taken to take 0, T for image, semantic class label vector and at the corresponding semantic classes target of appearance For transposition symbol；
Step 304: the 0th grade of global pool information being fully retained, other pyramids for having carried out information cover etc. are grading Mobile state competition, calculation formula are as follows:
Wherein, lclsFor loss function of always classifying, N is pyramid number of levels, l0For the damage of global pool pyramid grade It loses；
Further, above-mentioned step four specifically: error is calculated with loss function of always classifying and is calculated by backpropagation Method adjusts network parameter after the opposite variation of the loss function is less than 5%, directly by FsegIt takes out and is taken in classification dimension Prediction segmentation figure can be obtained in argmax
Y=argmax (Fseg)
Wherein, argmax FsegThe 3rd dimension.
The present embodiment verifies the validity of the method for the invention by Comparative result that image, semantic is divided.
Step 1 chooses DeepLab v2 network as depth characteristic extraction model, input pictureSource In PASCAL VOC semantic segmentation data set, the segmentation characteristic pattern handled by H
Step 2, segmentation characteristic pattern obtain characteristic of division figure by 1 × 1 convolution
Characteristic of division figure is carried out spatial pyramid pond by step 3, and pond core size is respectively 40,20,10,8 and 5, Obtain corresponding pyramid pond characteristic pattern With
Step 4 generates and covers tensor.It is 0 that ratio τ is covered in 0th grade of global pool region, and other pyramid hierarchical regions are covered Lid ratio τ and cover quantization parameter ρ is respectively 0.25 and 0.0.Tensor is so covered to meet:
The output characteristic pattern that each pyramid grade covers pond is calculated in step 5
Step 6 calculates classification activation vector:
Step 7: calculating class probability vector
Step 8, the intersection entropy loss for calculating each hierarchical pyramid,
Step 9 calculates competitive spatial pyramid cover pond loss function,
Step 10 repeats step 1 to step 9 progress network training to the more image samples of convolutional neural networks input, leads to Competitive spatial pyramid is crossed to cover pond loss function reverse propagated error and then update network.When test, one is inputted newly Image and extract segmentation characteristic patternPrediction segmentation figure Y is obtained after seeking argmax again.
It is noted that the setting of ratio τ and cover quantization parameter ρ that input image resolution, region are covered specifically are being flowed It is an example in journey, choosing also in the protected range of this patent in other situations.With average friendship and than (mIoU) As evaluation index, spatial pyramid cover pondization can by PASCAL VOC (Everingham, M., Van Gool, L., Williams,C.K.I.,Winn,J.and Zisserman, A.International Journal of Computer Vision, 88 (2), 303-338,2010) test set performance comparison such as the following table 1:
Table 1
Spatial pyramid cover pond is obvious to the promotion of performance, and accuracy rate alreadys exceed PASCAL VOC submission The full monitor model FCN-8s of list.Spatial pyramid cover pondization and global pool metaplasia at prediction segmentation result comparison such as Fig. 4. Multiple groups test comparative descriptions spatial pyramid and cover pondization with Small object identification, semantic relation correction and fringe region optimization Ability.In conclusion it is to have for the improvement that Weakly supervised image, semantic is divided that spatial pyramid proposed by the present invention, which covers pond, Effect.

Claims (7)

1. a kind of Weakly supervised image, semantic dividing method for covering pond based on spatial pyramid, comprising the following steps:
Step 1: selecting a convolutional neural networks H, handles input picture X by convolutional neural networks H, obtains characteristic of division Figure；
Step 2: establishing spatial pyramid pond module according to characteristic of division figure, then carries out spatial pyramid cover, obtains defeated Characteristic pattern out；
Step 3: calculating classification according to output characteristic pattern and activate vector sum class probability vector, then establishes competitive space gold Word tower covers pond loss function；
Step 4: covering pond loss function training convolutional neural networks H according to competitive spatial pyramid and extracts segmentation spy Sign figure.
2. a kind of Weakly supervised image, semantic dividing method that pond is covered based on spatial pyramid according to claim 1, It is characterized in that the step of handling input picture X by convolutional neural networks H in the step 1 specifically:
Step 101: selecting a convolutional neural networks H, map to obtain segmentation characteristic pattern by the convolutional neural networks HWherein, C is target category quantity, and W is weight parameter, For real number field；
Step 102: characteristic pattern dimensionality reduction will be divided by 1 × 1 convolution, obtain characteristic of division figureCalculation formula Are as follows:
Fcls=Fseg×W+b
Wherein, b is biasing,
3. a kind of Weakly supervised image, semantic dividing method that pond is covered based on spatial pyramid according to claim 1, It is characterized in that the step two specifically:
Step 201: average pond being carried out to all characteristic of division figures, while specifying total pyramid number and pond core size, is obtained To corresponding pyramid pond characteristic pattern Pi, wherein i be pyramid serial number, value range between 1 to total pyramid number it Between, calculate i-th of pyramid pond characteristic patternCalculation formula are as follows:
Wherein, kiFor i-th of pyramidal pond core size, x is characterized figure abscissa, and y is characterized figure ordinate, △ kxFor pond Change core abscissa, △ kyFor pond core ordinate, c is semantic classes quantity；
Step 202: each pyramid pond characteristic pattern being generated and covers tensorIt concurrently sets region and covers ratio Example and cover quantization parameter；
Step 203: according to the output characteristic pattern for covering tensor computation spatial pyramid cover pondIt calculates public Formula are as follows:
Oi=Pi⊙Mi
Wherein, ⊙ indicates hadamard product.
4. a kind of Weakly supervised image, semantic dividing method that pond is covered based on spatial pyramid according to claim 3, It is characterized in that remembering that i-th of pyramid covers tensor and be in the step 202Remember that jth pyramid-like is covered Tensor isIt is consistent with quantization parameter is covered that position is covered in the region that the different classes of pyramid covers tensor:
The ratio that the cover region generated at random accounts for the pyramid pond number of regions is τ,
Wherein, the value range of τ is [0,1], and the value of τ is bigger, and the region for indicating to cover is more,It indicates (1- τ) HiThe value that W is obtained is rounded downwards, and I () indicates that indicator function, I () take 1 when input condition meets, take 0 in other situations, I.e.The position of satisfaction is to be retained region, and other situations are the region covered；
The cover quantization parameter ρ is for inhibiting signal；
Wherein, the value range of ρ is [0,1].
5. a kind of Weakly supervised image, semantic dividing method that pond is covered based on spatial pyramid according to claim 1, It is characterized in that calculating classification in the step 3 activates vector sum class probability vector specifically:
Step 301: calculating classification and activate vector oi, i.e., the output characteristic pattern that spatial pyramid covers pond is classified as a value, counted Calculate formula are as follows:
Step 302: class probability vector being calculated by Softmax function, obtains the probability value of each semantic classes to be determined, is counted Calculate formula are as follows:
6. a kind of Weakly supervised image, semantic dividing method that pond is covered based on spatial pyramid according to claim 1, It is characterized in that in the step 3, establish competitive spatial pyramid cover pond loss function specifically includes the following steps:
Step 303: calculating the Classification Loss l of each pyramid gradei, calculation formula are as follows:
Wherein,Taking 1 other places to take 0, T for image, semantic class label vector and at the corresponding semantic classes target of appearance is to turn Set symbol；
Step 304: the 0th grade of global pool information being fully retained, other pyramid grades for having carried out information cover are moved State competition, calculation formula are as follows:
Wherein, lclsFor loss function of always classifying, N is pyramid number of levels, l0For the loss of global pool pyramid grade；
7. a kind of Weakly supervised image, semantic dividing method that pond is covered based on spatial pyramid according to claim 1, It is characterized in that the step four specifically: calculate error with loss function of always classifying and adjust net by back-propagation algorithm Network parameter is after the opposite variation of the loss function is less than 5%, directly by FsegIt takes out and takes argmax that can obtain in classification dimension To prediction segmentation figure
Y=argmax (Fseg)
Wherein, argmax FsegThe 3rd dimension.
CN201810739297.1A 2018-07-06 2018-07-06 Weak supervision image semantic segmentation method based on spatial pyramid covering pooling Active CN109215034B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810739297.1A CN109215034B (en) 2018-07-06 2018-07-06 Weak supervision image semantic segmentation method based on spatial pyramid covering pooling

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810739297.1A CN109215034B (en) 2018-07-06 2018-07-06 Weak supervision image semantic segmentation method based on spatial pyramid covering pooling

Publications (2)

Publication Number Publication Date
CN109215034A true CN109215034A (en) 2019-01-15
CN109215034B CN109215034B (en) 2021-09-21

Family

ID=64989960

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810739297.1A Active CN109215034B (en) 2018-07-06 2018-07-06 Weak supervision image semantic segmentation method based on spatial pyramid covering pooling

Country Status (1)

Country Link
CN (1) CN109215034B (en)

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110472669A (en) * 2019-07-22 2019-11-19 华北电力大学（保定） A kind of image classification method
CN110517971A (en) * 2019-09-17 2019-11-29 集美大学 The method for evaluating monocrystalline silicon battery surface pyramid texture uniformity
CN110619369A (en) * 2019-09-23 2019-12-27 常熟理工学院 Fine-grained image classification method based on feature pyramid and global average pooling
CN110866550A (en) * 2019-11-01 2020-03-06 云南大学 Convolutional neural network, pyramid strip pooling method and malicious software classification method
EP3687152A1 (en) * 2019-01-23 2020-07-29 StradVision, Inc. Learning method and learning device for pooling roi by using masking parameters to be used for mobile devices or compact networks via hardware optimization, and testing method and testing device using the same
CN111627055A (en) * 2020-05-07 2020-09-04 浙江大学 Scene depth completion method based on semantic segmentation
CN111860173A (en) * 2020-06-22 2020-10-30 中国科学院空天信息创新研究院 Remote sensing image ground feature element extraction method and system based on weak supervision
CN111967479A (en) * 2020-07-27 2020-11-20 广东工业大学 Image target identification method based on convolutional neural network idea
CN112164065A (en) * 2020-09-27 2021-01-01 华南理工大学 Real-time image semantic segmentation method based on lightweight convolutional neural network
CN112927310A (en) * 2021-01-29 2021-06-08 上海工程技术大学 Lane image segmentation method based on lightweight neural network
CN113111916A (en) * 2021-03-15 2021-07-13 中国科学院计算技术研究所 Medical image semantic segmentation method and system based on weak supervision
CN113111711A (en) * 2021-03-11 2021-07-13 浙江理工大学 Pooling method based on bilinear pyramid and spatial pyramid
CN116503603A (en) * 2023-05-16 2023-07-28 北京交通大学 Training method of inter-class shielding target detection network model based on weak supervision semantic segmentation and feature compensation

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105956532A (en) * 2016-04-25 2016-09-21 大连理工大学 Traffic scene classification method based on multi-scale convolution neural network
CN107273868A (en) * 2017-06-28 2017-10-20 电子科技大学 A kind of method that the dump and coal gangue area of coal field are distinguished in remote sensing images
CN107644426A (en) * 2017-10-12 2018-01-30 中国科学技术大学 Image, semantic dividing method based on pyramid pond encoding and decoding structure
CN107784122A (en) * 2017-11-22 2018-03-09 殷周平 A kind of instance-level image search method represented based on multilayer feature
CN107945185A (en) * 2017-11-29 2018-04-20 北京工商大学 Image partition method and system based on wide residual pyramid pond network

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105956532A (en) * 2016-04-25 2016-09-21 大连理工大学 Traffic scene classification method based on multi-scale convolution neural network
CN107273868A (en) * 2017-06-28 2017-10-20 电子科技大学 A kind of method that the dump and coal gangue area of coal field are distinguished in remote sensing images
CN107644426A (en) * 2017-10-12 2018-01-30 中国科学技术大学 Image, semantic dividing method based on pyramid pond encoding and decoding structure
CN107784122A (en) * 2017-11-22 2018-03-09 殷周平 A kind of instance-level image search method represented based on multilayer feature
CN107945185A (en) * 2017-11-29 2018-04-20 北京工商大学 Image partition method and system based on wide residual pyramid pond network

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
A. CHAUDHRY, AT EL.: ""Discovering Class-Specific Pixels for Weakly-Supervised Semantic Segmentation"", 《ARXIV》 *
KAIMING HE,AT EL.: ""Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition"", 《IEEE TRANSACIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 *
李明威: ""图像分类中的卷积神经网络方法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3687152A1 (en) * 2019-01-23 2020-07-29 StradVision, Inc. Learning method and learning device for pooling roi by using masking parameters to be used for mobile devices or compact networks via hardware optimization, and testing method and testing device using the same
CN110472669A (en) * 2019-07-22 2019-11-19 华北电力大学（保定） A kind of image classification method
CN110472669B (en) * 2019-07-22 2021-07-23 华北电力大学（保定） Image classification method
CN110517971A (en) * 2019-09-17 2019-11-29 集美大学 The method for evaluating monocrystalline silicon battery surface pyramid texture uniformity
CN110619369B (en) * 2019-09-23 2020-12-11 常熟理工学院 Fine-grained image classification method based on feature pyramid and global average pooling
CN110619369A (en) * 2019-09-23 2019-12-27 常熟理工学院 Fine-grained image classification method based on feature pyramid and global average pooling
CN110866550A (en) * 2019-11-01 2020-03-06 云南大学 Convolutional neural network, pyramid strip pooling method and malicious software classification method
CN110866550B (en) * 2019-11-01 2022-06-14 云南大学 Convolutional neural network, pyramid strip pooling method and malicious software classification method
CN111627055A (en) * 2020-05-07 2020-09-04 浙江大学 Scene depth completion method based on semantic segmentation
CN111627055B (en) * 2020-05-07 2023-11-24 浙江大学 Scene depth completion method combining semantic segmentation
CN111860173B (en) * 2020-06-22 2021-10-15 中国科学院空天信息创新研究院 Remote sensing image ground feature element extraction method and system based on weak supervision
CN111860173A (en) * 2020-06-22 2020-10-30 中国科学院空天信息创新研究院 Remote sensing image ground feature element extraction method and system based on weak supervision
CN111967479A (en) * 2020-07-27 2020-11-20 广东工业大学 Image target identification method based on convolutional neural network idea
CN112164065A (en) * 2020-09-27 2021-01-01 华南理工大学 Real-time image semantic segmentation method based on lightweight convolutional neural network
CN112164065B (en) * 2020-09-27 2023-10-13 华南理工大学 Real-time image semantic segmentation method based on lightweight convolutional neural network
CN112927310A (en) * 2021-01-29 2021-06-08 上海工程技术大学 Lane image segmentation method based on lightweight neural network
CN112927310B (en) * 2021-01-29 2022-11-18 上海工程技术大学 Lane image segmentation method based on lightweight neural network
CN113111711A (en) * 2021-03-11 2021-07-13 浙江理工大学 Pooling method based on bilinear pyramid and spatial pyramid
CN113111916A (en) * 2021-03-15 2021-07-13 中国科学院计算技术研究所 Medical image semantic segmentation method and system based on weak supervision
CN116503603A (en) * 2023-05-16 2023-07-28 北京交通大学 Training method of inter-class shielding target detection network model based on weak supervision semantic segmentation and feature compensation
CN116503603B (en) * 2023-05-16 2024-02-23 北京交通大学 Training method of inter-class shielding target detection network model based on weak supervision semantic segmentation and feature compensation

Also Published As

Publication number Publication date
CN109215034B (en) 2021-09-21

Similar Documents

Publication Publication Date Title
CN109215034A (en) A kind of Weakly supervised image, semantic dividing method for covering pond based on spatial pyramid
CN105184309B (en) Classification of Polarimetric SAR Image based on CNN and SVM
CN108717568B (en) A kind of image characteristics extraction and training method based on Three dimensional convolution neural network
CN108830242A (en) SAR image targets in ocean classification and Detection method based on convolutional neural networks
CN104732243B (en) SAR target identification methods based on CNN
CN106023065A (en) Tensor hyperspectral image spectrum-space dimensionality reduction method based on deep convolutional neural network
CN104217214B (en) RGB D personage's Activity recognition methods based on configurable convolutional neural networks
CN109508710A (en) Based on the unmanned vehicle night-environment cognitive method for improving YOLOv3 network
CN107871136A (en) The image-recognizing method of convolutional neural networks based on openness random pool
CN110503112A (en) A kind of small target deteection of Enhanced feature study and recognition methods
CN109785344A (en) The remote sensing image segmentation method of binary channel residual error network based on feature recalibration
CN109614985A (en) A kind of object detection method based on intensive connection features pyramid network
CN106682697A (en) End-to-end object detection method based on convolutional neural network
CN108734171A (en) A kind of SAR remote sensing image ocean floating raft recognition methods of depth collaboration sparse coding network
CN107909015A (en) Hyperspectral image classification method based on convolutional neural networks and empty spectrum information fusion
CN106504064A (en) Clothes classification based on depth convolutional neural networks recommends method and system with collocation
CN104408481B (en) Classification of Polarimetric SAR Image method based on depth wavelet neural network
CN107229904A (en) A kind of object detection and recognition method based on deep learning
CN107527352A (en) Remote sensing Ship Target contours segmentation and detection method based on deep learning FCN networks
CN107563355A (en) Hyperspectral abnormity detection method based on generation confrontation network
CN106682569A (en) Fast traffic signboard recognition method based on convolution neural network
CN106600595A (en) Human body characteristic dimension automatic measuring method based on artificial intelligence algorithm
CN112241679B (en) Automatic garbage classification method
CN107358258A (en) SAR image target classification based on the double CNN passages of NSCT and Selective Attention Mechanism
CN105005789B (en) A kind of remote sensing images terrain classification method of view-based access control model vocabulary

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant